International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022
p-ISSN: 2395-0072
www.irjet.net
Book Recommendation System Using Deep Learning (GPT3) Harsh Dubey1, Dr. Suma Kamalesh Gandhimathi2 Under The Guidance of, Dr. Suma Kamalesh Gandhimathi, Assistant Professor, Lovely Professional University, Punjab, School Of Computer Science and Engineering ---------------------------------------------------------------------------***------------------------------------------------------------------------Abstract Books have been an important part throughout different phases of peoples’ life. They serve as a source of knowledge, entertainment, stress relief, and most importantly motivation to do better in one’s life. Even though books are so important and choosing the correct book can be the difference between having a good and productive time reading it to getting frustrated at the halfway mark over the unnecessary expense made at purchasing that book and the time lost over getting to that point, from getting to actually learn something and get inspired to work towards your goals to getting back to square one i.e., going to the book-choosing step. This is where recommender systems step in. The main aim of this system from the very beginning has been to provide a simple, minimalistic yet highly functional, accurate, and information-rich interface for the user. This book recommender not only returns with book names and their cover page but also tries to embed various other useful information which can be the deciding factor in choosing a correct book, information such as a small description, preview of the book (if available), number of pages, etc. Keywords: Book Recommendation System, GPT3, Google Books, Streamlit, Python, Deep Learning
Introduction Recommender Systems have been a successful way of tackling issues where a suggestion is required for the viewer, listener, or reader. They even assist buyers by suggesting the products that they might be interested in. Through this project we have tried to create a recommendation system that will help readers by suggesting a book that they have already read and would like to read a similar book. Many recommender systems take user ratings into account for suggesting a book for the user but then those books are not highly related to the book that customer read, they are the books that other users who read the same book liked. Booksellers and libraries often struggle with space for books and with correct categorization of the book as they get many books which are not even in demand for their readers. This Book Recommendation System using Deep learning (GPT3) project has been an effort to take all these situations into account and then search for a better solution. Much like every other recommendation system available online for different tasks, this project serves as a tool to recommend books like what the user has already read and input its value into the form. This project can serve as the main tool for the customer, shopkeeper, librarian, and the people visiting the library to help them suggest, search, organize and even buy books.
Literature Review This recommender system uses GPT3 at its core and is assisted by python and streamlit to handle the creation of front-end and to establish the connection between front-end and the openai api along with the connection for google books api. GPT-n is developed, maintained, and offered by researchers at openai who came out with an idea that combining the transformer into a model which would be pretrained with a lot of data points and could be fine-tuned for later usage in specific models. It involves the usage of the transformer, which is an architecture used for transforming one type of sequence into any other with help of decoders and encoders. It also involves a self-attention mechanism which means that it weighs differently the significance of different parts of input data and can learn the context of actual sentences by using this method. GPT3 is the latest iteration of the generative pre-trained transformer. It is an autoregressive language model, and it can produce human-like texts using deep learning. It has trained over more than 1.75 billion data points and offers a lot of
© 2022, IRJET
|
Impact Factor value: 7.529
|
ISO 9001:2008 Certified Journal
|
Page 347